fp7 uniboard project
DESCRIPTION
Digital Receiver G. Comoretto, A. Russo, G. Tuccari, A Baudry, P. Camino, B. Quertier Dwingeloo, February 27, 2009. FP7 Uniboard project. Framework. Wideband radioastronomy receivers Instantaneous BW 4-12 GHz Digitizers: current 1-2 GHz, near future 4-8 GHz - PowerPoint PPT PresentationTRANSCRIPT
FP7 Uniboard projectFP7 Uniboard project
Digital Receiver
G. Comoretto, A. Russo, G. Tuccari, A Baudry, P. Camino, B. Quertier
Dwingeloo, February 27, 2009
FrameworkFramework Wideband radioastronomy receivers
Instantaneous BW 4-12 GHz Digitizers: current 1-2 GHz, near future 4-8
GHz Digital hardware: limited by clock, 100-300
MHz bandwidth→ Need for frequency multiplexing Multibeam receivers
Tens → hundreds of pixels
7 feed FP5 Faraday receiver
Existing systemsExisting systems DBBC
4x 1GHz BW, up to 16 Virtex5 cores EVN standard: 4 cores = VLBI terminal DBBC3: wider band ADC3 & V.6 core
MPI-Agilent FFT Spectrometer 1 GHz BW, Virtex4 core Up to 8 boards on a PCI bus
Casper Up to 1.5 GHz BW, 1x (IBOB) or 5 x
Virtex2 (BEE2) ROACH: Virtex-5 based
Avoid duplications!Keep an open path to future apps & HW
ROACH
DBBC
SpecificationsSpecifications Input:
Bandwidth: 4 GHz (8 GS/s) Quantization: 6 bit minimum (8 bit possible) Future development, 2-3 yrs: 8 GHz 4-6 bit Format: data on several 10G links, one sample per link, time
multiplexed Output:
Bandwidth: up to 64 MHz/channel (128 MS/s), up to 64 channels Quantization: 1, 2, 4, 8 bit Format: Standard MK5C format, UDP packets, on 10G links
Each channel position and BW independently selectable (with limitations)
Everything fits in a single board
General structureGeneral structure Two stage filter 1st stage:
Polyphase FFT real input 32x250MHz complex output, divided into 8 groups 125 MHz channel spacing
2nd stage: Tunable filter: LO/mixer, low-pass FIR, complex-to-real 4 sets of 16 FIRs Each set selects input among a 1st stage channel group
High SFDR using polyphase filterHigh SFDR using polyphase filter8 GHz BW likely affected by RFIRFI mitigation by excision of
affected spectral regionGood chan.-to-chan. rejection to
avoid contamination18 bit multipliers: limit to ~85 dB
SFDRFilter design: 85 dB stopband96% useful BW, 0.04 dB ripple>90 dB with 94% BW
channel 3
channel 2
channel 1
Channels overlapped to avoid holes
Out BW
How to fit in the UniboardHow to fit in the Uniboard
Uniboard structure: 4 input + 4 output
FPGAs Each FPGA has 2x10G
in/out No vertical interconnect 4x4 fixed interconnect
between input and output FPGAs
2.5 Gbps per link 250 MS/s @ 8 bit
2 stage DIT FFT2 stage DIT FFT
Decimation in time FFT algorithm: separate data flow for parallel decimated samples
Limitation in interconnections: only selected channels processed
Channels and groupsChannels and groups32 overlapping channels½ channel spacing
Channels are grouped in groups of 4Groups 0, 1, 2, 3
Groups 4, 5, 6, 7
Up to 4 groups available at each time:segmented coverage of whole 4 GHz (with edge holes, ~4% lost) or continuous coverage of narrower band
8 bit between stages enough?8 bit between stages enough?
Limited BW between input and output FPGAs: max 256MS @ 8 bit per link
Requantization with 8 bit may degrade SFDR
Tests with Gaussian noise + spectral line in 1 channel
> 80 dB SFDR observed
22ndnd stage FIR stage FIR
Derived from ALMA Tunable FilterbankVariable Decimation FIR Filter: D=2-256
Taps: 32*D, tap recirculation, symmetric
32 (2x16, real-imaginary) multipliersOutput conversion to real by ½ band shiftSpecifications TBD.
Example from ALMA design 18 bit required for high rejection Tradeoff between complexity and
performances
Example: filter performance for 9 bit, 94% BW, 16 multipliers½ to 1/256 BW
ConclusionsConclusions
4 GHz BW, 64 tunable channels system feasible in a single Uniboard
4x1GHz bypassing 2nd butterfly (but easier w. DBBC!) 8 GHz BW feasible (Phase 2 of the project) >80 dB spurious free dynamic range (may improve) Applications
Front-end for other systems VLBI channelization system FFT very wide band spectrometer (e.g. 4GHz 64k pts)